Bạn đang xem bản rút gọn của tài liệu. Xem và tải ngay bản đầy đủ của tài liệu tại đây (109.65 KB, 6 trang )
<span class='text_page_counter'>(1)</span><div class='page_container' data-page=1>
Math 598 Feb 2, 20051
Geometry and Topology II
Spring 2005, PSU
If<i>M</i> is a smooth <i>n</i>-dimensional manifold, then to each point <i>p</i>of<i>M</i> we may
associate an <i>n</i>-dimensional vector space<i>TpM</i> which is defined as follows. Let
<i>CurvespM</i> :=<i>{α</i>: (<i>−, </i>)<i>→M</i> <i>|α</i>(0) =<i>p}</i>
be the space of <i>smooth</i> curves on <i>M</i> centered at <i>p</i>. We say that a pair of
curve<i>α</i>,<i>β</i> <i>∈CurvespM</i> are <i>tangent</i> at<i>p</i>, and we write<i>α</i> <i>∼β</i>, provided that
there exists a local chart (<i>U, φ</i>) of <i>M</i> centered at <i>p</i>such that
(<i>φ◦α</i>)<i></i>(0) = (<i>φ◦β</i>)<i></i>(0)<i>.</i>
Note that if (<i>V, ψ</i>) is any other local chart of <i>M</i> centered at<i>p</i>, then, by the
chain rule,
(<i>ψ◦α</i>)<i></i>(0) = (<i>ψ◦φ−</i>1<i>◦φ◦α</i>)<i></i>(0)
=
(<i>ψ◦φ−</i>1)<i></i>(<i>φ</i>(<i>α</i>(0))
<i>◦</i>(<i>φ◦α</i>)<i></i>(0)
=
(<i>ψ◦φ−</i>1)<i></i>(<i>φ</i>(<i>β</i>(0))
<i>◦</i>(<i>φ◦β</i>)<i></i>(0)
= (<i>ψ◦β</i>)<i></i>(0)<i>.</i>
Thus<i>∼</i>is well-defined, i.e., it is indpendent of the choice of local coordinates.
Further, one may easily check that <i>∼</i> is an equivalence relation. The set of
<i>tangent vectors</i> of <i>M</i> at<i>p</i> is defined by
<i>TpM</i> :=<i>CurvespM/∼.</i>
Next we describe how<i>TpM</i> may be given the structure of a vector space.
Let (<i>U, φ</i>) denote, as always, a local chart of<i>M</i> centered at<i>p</i>, and recall that
<i>n</i> = dim(<i>M</i>). Then we define a mapping <i>f</i>: <i>TpM</i> <i>→</i><b>R</b><i>n</i> by
<b>Exercise 1.</b> Show that the above mapping is well-defined and is a bijection.
Since <i>φ<sub>∗</sub></i> is a bijection, we may use it to identify <i>TpM</i> with <b>R</b><i>n</i> and, in
particular, define a vector space structure on <i>TpM</i>. More explicitly, we set
[<i>α</i>] + [<i>β</i>] :=<i>φ−<sub>∗</sub></i>1<i>φ<sub>∗</sub></i>([<i>α</i>]) +<i>φ<sub>∗</sub></i>([<i>β</i>])<i>,</i>
and
<i>λ</i>[<i>α</i>] :=<i>φ−<sub>∗</sub></i>1<i>λφ<sub>∗</sub></i>([<i>α</i>])<i>.</i>
Here we give a more abstract, but useful, characterization for the tangent
space of a manifold, which reveals the intimate connection between tangent
vectors and directional derivatives.
Let<i>C∞</i>(<i>M</i>) denote the space of smooth functions on <i>M</i> and <i>p∈M</i>. We
say that two functions <i>f, g</i> <i>∈</i> <i>C∞</i>(<i>M</i>) have the same <i>germ</i> at <i>p</i>, and write
<i>f</i> <i>∼p</i> <i>g</i>, provided that there exists an open neighborhood <i>U</i> of <i>p</i> such that
<i>f|U</i> =<i>g|U</i>. The reslting equivalence classes then defines the space of germ of
smooth functions of <i>M</i> at<i>p</i>:
<i>Cp</i>(<i>M</i>) :=<i>C∞M/∼p</i> <i>.</i>
Note that we can add and multiply the elements of <i>CpM</i> in an obvious way,
and with respect to these operations one may easily check that <i>Cp</i>(<i>M</i>) is an
<i>algebra</i> over the field of real numbers<b>R</b>.
We say that a mapping<i>D</i>: <i>Cp</i>(<i>M</i>)<i>→</i><b>R</b> is a <i>derivation</i> provided that <i>D</i>
is linear and satisfies the Leibnitz rule, i.e.,
<i>D</i>(<i>f g</i>) = <i>Df·g</i>(<i>p</i>) +<i>f</i>(<i>p</i>)<i>·Dg</i>
for all <i>f</i>, <i>g</i> <i>∈</i> <i>Cp</i>(<i>M</i>). If <i>D</i>1 and <i>D</i>2 are a pair of such derivations, then we
define their sum by (<i>D</i>1+<i>D</i>2)<i>f</i> :=<i>D</i>1<i>f</i>+<i>D</i>2<i>f</i>, and for any<i>λ∈</i><b>R</b>, the scalar
product is given (<i>λD</i>)<i>f</i> :=<i>λ</i>(<i>Df</i>).
<b>Exercise 2.</b> Show that the set of derivations of <i>CpM</i> forms a vector space
with respect to the operations defined above.
Note that each element <i>X</i> <i>∈</i> <i>TpM</i> gives rise to a derivation of <i>Cp</i>(<i>M</i>) if,
for any <i>f</i> <i>∈Cp</i>(<i>M</i>), we set
where <i>αX</i>: (<i>−, </i>) <i>→</i> <i>M</i> is a curve which belongs to the equivalence class
denoted by <i>X</i>, i.e., <i>X</i> = [<i>αX</i>].
<b>Exercise 3.</b> Check that<i>Xf</i> is well-defined and is indeed a derivation.
A much less obvious fact, whose demonstrationis the main aim of this
section, is that, conversely, every derivation of <i>Cp</i>(<i>M</i>) corresponds to (the
directional derivative determined by) a tangent vecor. More formally, if<i>DpM</i>
denotes the space of derivations of <i>CpM</i>, then
<b>Theorem 4.</b> <i>TpM</i> <i>is isomorphic to</i> <i>DpM.</i>
The rest of this section is devoted to the proof of the above result. To this
end we need a pair of lemmas. Let <b>0</b><i>∈Cp</i>(<i>M</i>) denote the constant function
zero, i.e. <b>0</b>(<i>p</i>) := 0.
<b>Lemma 5.</b> <i>If</i> <i>f</i> <i>∈CpM</i> <i>is a constant function, then</i> <i>Df</i> =<b>0</b><i>, for any</i> <i>D</i> <i>∈</i>
<i>DpM.</i>
<i>Proof.</i> First note that, since <i>f</i> is constant, say <i>f</i>(<i>p</i>) = <i>λ</i>,
<i>D</i>(<i>f</i>) =<i>D</i>(<i>f</i> <i>·</i><b>1</b>) =<i>D</i>(<i>λ·</i><b>1</b>) =<i>λD</i>(<b>1</b>)<i>,</i>
where <b>1</b> denotes the constant fucntion <b>1</b>(<i>p</i>) = 1. Further,
<i>D</i>(<b>1</b>) = <i>D</i>(<b>1</b><i>·</i><b>1</b>) =<i>D</i>(<b>1</b>)<i>·</i>1 + 1<i>·D</i>(<b>1</b>) = 2<i>D</i>(<b>1</b>)<i>.</i>
Thus <i>D</i>(<b>1</b>) = 0, which in turn yields that <i>D</i>(<i>f</i>) = <b>0</b>.
<b>Lemma 6.</b> <i>Let</i> <i>f</i>: <b>R</b><i>n</i> <i><sub>→</sub></i> <b><sub>R</sub></b> <i><sub>be a smooth function. Then, for any</sub></i> <i><sub>p</sub></i> <i><sub>∈</sub></i> <b><sub>R</sub></b><i>n<sub>,</sub></i>
<i>there exist smooth functions</i> <i>gi</i><sub>:</sub> <b><sub>R</sub></b><i>n<sub>→</sub></i><b><sub>R</sub></b><i><sub>,</sub></i> <i><sub>i</sub></i><sub>= 1</sub><i><sub>, . . . , n</sub><sub>, such that</sub></i>
<i>gi</i>(<i>p</i>) = <i>∂f</i>
<i>∂xi</i>
(<i>p</i>)<i>,</i>
<i>and</i>
<i>f</i>(<i>x</i>) =<i>f</i>(<i>p</i>) +
<i>n</i>
<i>i</i>=1
<i>Proof.</i> The fundamental theorem of calculus followed by chain rule implies
that
<i>f</i>(<i>x</i>)<i>−f</i>(<i>p</i>) =
1
0
<i>d</i>
<i>dtf</i>(<i>tp</i>+ (1<i>−t</i>)<i>x</i>)<i>dt</i>
=
1
0
<i>n</i>
<i>i</i>=1
<i>∂f</i>
<i>tp</i>+(1<i><sub>−</sub>t</i>)<i>x</i>(<i>x</i>
<i>i</i> <i><sub>−</sub><sub>p</sub>i</i><sub>)</sub><i><sub>dt</sub></i>
=
<i>n</i>
<i>i</i>=1
1
0
<i>∂f</i>
<i>dxi</i>
<i>tp</i>+(1<i>−t</i>)<i>xdt</i>(<i>x</i>
<i>i<sub>−</sub></i>
<i>pi</i>)<i>.</i>
So we set
<i>gi</i>(<i>x</i>) :=
1
0
<i>tp</i>+(1<i>−t</i>)<i>xdt.</i>
Now we are ready to prove the main result of this section
<i>Proof of Theorem 4.</i> Recall that if (<i>U, φ</i>) is a local chart of <i>M</i> centered at<i>p</i>,
then the mapping [<i>α</i>]<i>→</i>(<i>φ◦α</i>)<i></i>(0) is an isomorphism between<i>TpM</i> and <b>R</b><i>n</i>.
Similarly, <i>f</i> <i>→</i> <i>f</i> <i>◦φ−</i>1 <sub>is an isomorphism between</sub> <i><sub>C</sub></i>
<i>pM</i> and <i>Co</i><b>R</b><i>n</i>, which
yields that <i>DpM</i> is isomorphic to <i>Do</i><b>R</b><i>n</i>. So it remains to show that <i>Do</i><b>R</b><i>n</i>
is isomorphic to <b>R</b><i>n</i>.
Let<i>xi</i><sub>:</sub><b><sub>R</sub></b><i>n</i> <i><sub>→</sub></i><b><sub>R</sub></b><sub>, given by</sub><i><sub>x</sub>i</i><sub>(</sub><i><sub>p</sub></i><sub>) :=</sub><i><sub>p</sub>i</i><sub>, be the coordinate functions of</sub> <b><sub>R</sub></b><i>n</i><sub>.</sub>
It is easy to check that the mapping
<i>Do</i><b>R</b><i>nD</i>
<i>F</i>
<i>−→</i>(<i>Dx</i>1<i>, . . . , Dxn</i>)<i>∈</i><b>R</b><i>n</i>
is a homomorphism. Furhter, <i>F</i> is one-to-one because, by the previous
<i>Df</i> = 0 +
<i>n</i>
<i>i</i>=1
(<i>Dgi·xi</i>(<i>o</i>) +<i>gi</i>(<i>o</i>)<i>·Dxi</i>) =
<i>n</i>
<i>i</i>=1
<i>∂f</i>
<i>∂xi</i>
(<i>o</i>)<i>Dxi.</i>
In particular, knowledge of<i>Dxi</i> <sub>uniquely determines</sub><i><sub>D</sub></i><sub>. Finally it remains to</sub>
show that <i>F</i> is onto. To this end note that to each <i>X</i> = (<i>X</i>1<i><sub>, . . . , X</sub>n</i><sub>)</sub><i><sub>∈</sub></i><b><sub>R</sub></b><i>n</i><sub>,</sub>
we may assign a derivation of <i>Cp</i><b>R</b><i>n</i> given by
<i>DX</i> :=
<i>i</i>=1
<i>Xi</i> <i>∂</i>
<i>∂xi</i>
<i>x</i>=<i>o.</i>
<b>Exercise 7.</b> Show that any local chart (<i>U, φ</i>) of<i>M</i> centered at<i>p</i>determines
a basis <i>E</i><sub>1</sub><i>φ, . . . Eφ</i>
<i>n</i> for <i>TpM</i> as follows. For every <i>f</i> <i>∈CpM</i>, set:
<i>E<sub>i</sub>φf</i> := <i>∂</i>(<i>f◦φ</i>
<i>−</i>1<sub>)</sub>
<i>∂xi</i>
(<i>o</i>)<i>.</i>
Let <i>f</i>: <i>M</i> <i>→N</i> be a smooth map, and <i>p∈M</i>. Then the <i>differential</i> of <i>f</i> at
<i>p</i> is the mapping <i>dfp</i>: <i>TpM</i> <i>→Tf</i>(<i>p</i>)<i>N</i> given by
<i>dfp</i>([<i>α</i>]) := [<i>f</i> <i>◦α</i>]<i>.</i>
<b>Exercise 8.</b> Show that if <i>f</i>: <b>R</b><i>n</i> <i><sub>→</sub></i> <b><sub>R</sub></b><i>m</i> <sub>and we identify</sub> <i><sub>T</sub></i>
<i>p</i><b>R</b><i>n</i> and <i>Tf</i>(<i>p</i>)<b>R</b><i>m</i>
with<b>R</b><i>n</i>and<b>R</b><i>m</i>respectively in the standard way (i.e., via the mapping [<i>α</i>]<i>→</i>
<i>α</i>(0)) then <i>dfp</i> may be identified with the linear transformation determined
by the jacobian matrix (<i>∂fi<sub>/∂x</sub></i>
<i>j</i>) (in particular,<i>dfp</i> is a generalization of the
standard derivative <i>Df</i>(<i>p</i>) of maps between Euclidean spaces).
Using the characterization of <i>TpM</i> as the space of derivations over the
germ of smooth functions of <i>M</i> at <i>p</i>, one may give an alternative definition
of <i>dfp</i> as follows. Given <i>X</i> <i>∈TpM</i>, we define
<i>dfp</i>(<i>X</i>) <i>g</i> :=<i>X</i>(<i>g◦f</i>)<i>,</i>
for any <i>g</i> <i>∈Cf</i>(<i>p</i>)<i>N</i>. Thus <i>dfp</i>(<i>X</i>)<i>∈Df</i>(<i>p</i>)<i>N</i> <i>Tf</i>(<i>p</i>)<i>N</i>. Note that if <i>X</i> = [<i>α</i>],
then
<i>X</i>(<i>g◦f</i>) = (<i>g◦f◦α</i>)<i></i>(0) = [<i>f◦α</i>]<i>g.</i>
Thus the two definitions of <i>dfp</i> presented above are indeed equivalent. Using
the second definition, one may immediately check that <i>dfp</i> is a
homomor-phism. Another fundamental property is:
<b>Exercise 9 (The chain rule).</b> Show that if<i>f</i>: <i>M</i> <i>→N</i> and<i>g</i>: <i>N</i> <i>→L</i>are
smooth maps, then, for any <i>p∈M</i>,
<i>d</i>(<i>g◦f</i>)<i>p</i> =<i>dgf</i>(<i>p</i>)<i>◦dfp.</i>
We say <i>f</i>: <i>M</i> <i>→</i> <i>N</i> is a <i>diffeomorphism</i> if <i>f</i> is a homeomorphism, and
<i>f</i> and <i>f−</i>1 <sub>are smooth. If there exists a diffeomorphism between a pair of</sub>
<b>Exercise 10.</b> Show that if <i>f</i>: <i>M</i> <i>→</i> <i>N</i> is a diffeomorphism, then <i>dfp</i> is an
isomorphism for all <i>p</i> <i>∈</i> <i>M</i>. In particular, conclude that if <i>M</i> and <i>N</i> are
diffeomorphic, then <i>dim</i>(<i>M</i>) = <i>dim</i>(<i>N</i>).
Note that the last statement if the above exercise also follows from the
standard fact in Algebraic topology that<b>R</b><i>n</i> <sub>and</sub> <b><sub>R</sub></b><i>m</i> <sub>are homemorphic only</sub>